A trading strategy can become less reliable the more you try to perfect it. Imagine testing a strategy on 20 years of historical data. You notice that certain losses occurred before major reversals. So you add a rule “Don’t take the trade when this pattern appears.” The results improve. Then you find another losing pattern. Add another rule. The backtest improves again. Repeat this enough times… And eventually you can build a strategy that looks exceptional on the past. But there is a problem. You may have optimized the strategy for the data you already saw. This is the basic danger of overfitting. A model can become extremely good at explaining historical observations while becoming less useful on new data. And trading makes this particularly dangerous because markets contain noise, changing regimes, and relatively few truly independent observations. The more decisions you make after seeing the data, the easier it becomes to mistake coincidence for an edge. So when a backtest gets dramatically better after adding one more rule, don’t only ask “Did the strategy improve?” Ask “Did I discover something real or did I teach the strategy the answer to an exam it has already seen?